On March 27, Zhang Yiming, ByteDance's founder, ascended to the top of China's wealth rankings, with ByteDance continuing its foray into embodied intelligence and initiating an aggressive "talent grab" strategy. The company aims to recruit 5,000 individuals during the spring hiring season, with 60% of the positions dedicated to R&D and a 30% surge in demand for non-R&D roles. Vacancies encompass natural language processing, computer vision, machine learning, and robotics.

In embodied intelligence, ByteDance's latest R&D positions predominantly focus on robotic "algorithms," offering around 10 such roles.

Among them, the SLAM deep learning algorithm engineer will be responsible for researching and developing robot localization technology, as well as localization algorithms. The robot motion control algorithm engineer will develop algorithms for multi-degree-of-freedom robotic arm task decision-making, motion planning, and collaborative control, alongside algorithms for robot safety, autonomous obstacle avoidance, and grasping operations. Lastly, the robot sensor system engineer will handle the embedded software and hardware development of sensors in robotics.

The "talent gap" in embodied intelligence is a ubiquitous challenge for all industry players. According to reports, the overall talent shortage in China's AI sector stands at an astonishing 5 million. As an avant-garde domain within AI, embodied intelligence faces an even more acute talent deficit, particularly for high-precision technical and composite talents. By 2025, Shanghai anticipates a demand for 400,000 embodied intelligence talents, whereas the number of relevant majors in local universities stands at merely 40,000.

Against this backdrop, the urgency of "talent grabbing" intensifies. Only by swiftly amassing top-tier technical talents and accelerating technology R&D and product iteration can companies establish an early foothold in embodied intelligence and seize the strategic high ground of the industry.

Concurrently, ByteDance's investment in computing power is escalating. In 2023, the company ordered GPUs worth over $1 billion from NVIDIA. In May 2024, ByteDance announced a tender for its self-built East China data center, scheduled for completion and operation in 2025. Additionally, plans for constructing large data centers in Guangdong and other locations are underway, providing robust computing power support for ByteDance's embodied intelligence endeavors.

Having fortified its team and enhanced software capabilities, ByteDance now stands on a solid foundation to further advance in embodied intelligence, leveraging its formidable computing power and advanced algorithms.

Public records indicate that ByteDance began investing in robot companies like CloudMinds in 2019 and transitioned to internal R&D of embodied intelligence in 2020.

In July 2023, ByteDance announced its entry into the robotics industry, initiating team expansion by year-end. By 2024, the company's robotics team, FAE, neared 100 members, with the AMR (mobile robot) team comprising two-thirds and general embodied intelligence accounting for one-third.

FAE is currently part of ByteDance's AILab, led by Director Li Hang, a former director and chief scientist at Huawei's Noah's Ark Lab. Li Hang reports to Yang Zhenyuan, ByteDance's vice president and head of algorithm technology.

AILab, established in 2016 by an internal team, comprises multiple sub-teams focusing on computer vision, natural language processing, speech and audio processing, and machine learning.

With a focus on humanoid robot "brain" research, ByteDance enjoys a first-mover advantage in technology R&D, algorithm capabilities, technology iteration and innovation, and data-driven algorithm optimization. Leveraging vast internet data and interaction data from real-world robot clusters, ByteDance can swiftly advance technology implementation and maintain a leading position amidst competition with major players.

In November 2023, ByteDance unveiled the RoboFlamingo embodied operation large model, co-developed with Tsinghua University. This model employs a pre-trained visual-language model (VLM) for one-step visual-language understanding, fine-tuned through imitation learning, significantly enhancing robot operation generalization and success rates.

In October 2024, ByteDance introduced the second-generation large robot model, GR-2. Trained on 38 million internet video clips, GR-2 has initially acquired world model capabilities. In multi-task learning tests, GR-2 completed 105 diverse desktop tasks with an average success rate of 97.7%.

As major companies enter this space, the embodied intelligence sector is experiencing rapid growth. Alibaba has established Ant Lingbo Technology, aiming to develop industry-leading robots for the service sector. JD.com's embodied intelligence strategy focuses on household scenarios, with a dedicated business department already in place. vivo's Robot Lab studio concentrates on technological breakthroughs related to household robots' "brains" and "eyes." 2025 could mark the "first year" of humanoid robot evolution, with technological advancements and commercialization expected to accelerate.

Note: The lead image is sourced from ByteDance's official website.